From a number of GWASs, we selected 4,259 individuals who were not knowingly related to one another and confirmed this with SNP data. We then estimated their pairwise genetic relationships using all autosomal markers (MAF ≥ 0.01), and retained 3,925 individuals (3,248 adults and 677 16-year-olds) whose pairwise relationship was estimated at less than 0.025 (maximum relatedness approximately corresponding to cousins two to three times removed: Supplementary Fig. 1). We fitted a linear model to the height data and used restricted maximum likelihood (REML)21 to estimate the variance explained by the SNPs. (In Online Methods, we show how this can be conveniently implemented with a mathematically equivalent model that uses the SNPs to calculate the genomic relationship between pairs of subjects). Using this approach, we estimated the proportion of phenotypic variance explained by the SNPs as 0.45 (s.e. = 0.08, Table 1), a nearly tenfold increase relative to the 5% explained by published and validated individual SNPs.
